PHZ 7427 SOLID STATE II: Electron-electron interaction and the Fermi-liquid theory

Transcription

1 PHZ 7427 SOLID STATE II: Electron-electron interaction and the Fermi-liquid theory D. L. Maslov Department of Physics, University of Florida (Dated: February 2, 204)

2 CONTENTS I.Notations 2 II.Electrostatic screening 2 A.Thomas-Fermi model 2 B.Effective strength of the electron-electron interaction. Parameter r s. 2 C.Full solution (Lindhard function) 3 D. Lindhard function 5.A discourse: properties of Fourier transforms 6 2.End of discourse 7 E.Friedel oscillations 7.Enhancement of the backscattering probability due to Friedel oscillations 8 F.Hamiltonian of the jellium model 9 G. Effective mass near the Fermi level 3.Effective mass in the Hartree-Fock approximation 5 2.Beyond the Hartree-Fock approximation 5 III. Stoner model of ferromagnetism in itinerant systems 6 IV. Wigner crystal 20 V.Fermi-liquid theory 22 A.General concepts 22.Motivation 23 B. Scattering rate in an interacting Fermi system 24 C.Quasi-particles 27.Interaction of quasi-particles 29 D.General strategy of the Fermi-liquid theory 3 E.Effective mass 3 F.Spin susceptibility 33.Free electrons 33 2.Fermi liquid 34 G. Zero sound 35 2

3 VI. Non- Fermi-liquid behaviors 37 A.Dirty Fermi liquids 38.Scattering rate 38 2.T-dependence of the resistivity 40 A.Born approximation for the Landau function 43 References 44 I. NOTATIONS k B = (replace T by k B T in the final results) = (momenta and wave numbers have the same units, so do frequency and energy) ν(ε) density of states II. ELECTROSTATIC SCREENING A. Thomas-Fermi model For Thomas-Fermi model, see AM, Ch. 7. B. Effective strength of the electron-electron interaction. Parameter r s. is The ratio of the Coulomb energy at a typical inter-electron distance to the Fermi energy r is found from U C E F = U C E F = e2 / r E F. ( ) / π r 3 n = r = n /3 4π ( ) /3 4π ( ) e2 n /3 m = 2 (2) 2/3 e 2 m 3 (3π 2 ) 2/3 /2 n 2/3 3 π n =. m /3 34e2 n. /3 Lower densities correspond to stronger effective interactions and vice versa. 3

4 Parameter r s is introduced as the average distance between electrons measured in units of the Bohr radius r = r s a B = r s /me 2. Expressing r s in terms of n and relating density to k F, we find r s = ( ) /3 3 = 4π n /3 a B ( ) /3 9π me 2. 4 k F In terms of r s, and U C = e2 r s a B E F = ( ) 2/3 9π. 2 4 ma 2 B rs 2 ( ) 2/3 U C 4 = 2 r s. 54r s. E F 9π C. Full solution (Lindhard function) In the Thomas-Fermi model, one makes two assumptions: a) the effective potential acting on electrons is weak and b) the effective potential (and corresponding density) varies slowly on the scale of the electron s wavelength. Assumption a) allows one to use the perturbation theory whereas assumption b) casts this theory into a quasi-classical form. In a full theory, one discards assumption b) but still keeps assumption a). So now we want to do a complete quantum-mechanical (no quasi-classical assumptions) form. Let the total electrostatic potential acting on an electron be φ = φ ext + φ ind, where φ ext is the potential of external charges and φ ind is that of induced charges. Correspondingly, the potential energy v = eφ = eφ ext eφ ind. Because we are doing the linear-response theory, the form of the external perturbation does not matter. Let s choose it as a plane-wave v ( r, t) = 2 v qe i( q r ωt) + c.c. () 4

5 Before the perturbation was applied, the wavefunction was Ψ 0 = L 3/2 e(ik r ε kt). The wavefunction in the presence of the perturbation is given by standard expression from the first-order perturbation theory Ψ = Ψ 0 [ + v q 2 e i( q r ωt) ε k ε k+ q + ω + v q 2 ] e i( q r ωt), ε k ε k q ω where the last term is a response to a c.c. term in Eq.(). The Fourier component of the wavefunction [ Ψ k = Ψ 0k + v q 2 ε k ε k+ q + ω + ] v q. 2 ε k ε k q ω The induced charge density is related to the wavefunction is ρ ind = 2e k f k ( Ψk 2 Ψ 0k 2), (2) where f k is the Fermi function, factor of 2 is from the spin summation and the homogeneous (unperturbed) charge density was subtracted off. Keeping only the first-order terms in v q, Eq.(2) gives ρ ind = 2e [ ] f L 3 k v q ε k ε k+ q + ω + ε k ε k q ω k d 3 k f k f k+ q = 2ev q (2π) 3 ε k ε k+ q + ω, (4) where we shifted the variables as k q k and k k + q in the last term. The charge susceptibility, χ, is defined as ρ ind = e q2 4π χv q = e 2 q2 4π χφ q, (5) where φ q is the Fourier component of the net electrostatic potential. Comparing Eqs. (4) and (5), we see that χ = 4π q ( 2) 2 d 3 k f k f k+ q (2π) 3 ε k ε k+ q + ω. The meaning of χ becomes more clear, if we write down the Poisson equation (in a Fouriertransformed form) q 2 φ q = 4π (ρ ext + ρ ind ). 5 (3)

6 External charges and potentials satisfy a Poisson equation on their own q 2 φ ext = 4πρ ext. Now, q 2 φ q = q 2 φ ext + 4πρ ind = q 2 φ ext 4πe 2 q2 4π χφ q φ q = φ ext + χe. 2 Using a definition of the dielectric function we see that φ q = ɛ (q, ω) = + χe 2 = + 4πe2 q 2 φ ext ɛ (q, ω), (6) ( 2) This is the Lindhard s expression for the dielectric function. d 3 k f k f k+ q (2π) 3 ε k ε k+ q + ω. (7) a. Check Let s make sure that the general form of ɛ (q, ω) [Eq.(7)] does reduce to the Thomas-Fermi one in the limit ω = 0 and q k F. For small q, and At T = 0, f ε k Now, ɛ (q, 0) = + 4πe2 q 2 ( 2) d 3 k f k f k+ q (2π) 3 ε k ε k+ q f k+ q = f (ε k+ q ) = f (ε k+ q ε k + ε k ) = f (ε k ) + f ε k (ε k+ q ε k ) +... ɛ (q, 0) = + 4πe2 q 2 2 ( d 3 k (2π) 3 f ). ε k = δ (ε k E F ). The density of states at the Fermi energy ν F = 2 d 3 k (2π) 3 δ (ε k E F ). ɛ (q, 0) = + 4πe2 q ν 2 F = + κ2 q, 2 where κ 2 4πe 2 ν F, which is just the Thomas-Fermi form. 6

7 D. Lindhard function As shown in AM, the static form of the Lindhard s dielectric function is given by [ ɛ (q, 0) = + 4πe2 q x2 4x ln + x ], x where x q/2k F. Notice that the derivative of ɛ (q, 0) is singular for q = 2k F, i.e., x =. This singularity gives rise to a very interesting phenomenon Friedel oscillations in the induced charge density (and corresponding potentials). Mathematically, it arises because of the property of the Fourier transform. To find the net electrostatic potential in the real space we need to Fourier transform back to real space Eq.(6). Let s say that the external perturbation is a single point charge Q. Then (in the q space), φ ext = 4πQ/q 2 and d 3 q 4πQ φ (r) = q 2 ɛ (q, 0). (2π) 3 e i q r (8). A discourse: properties of Fourier transforms Fourier transforms have the following property. Suppose we want to find the large t limit of dω F (t) = 2π e iωt F (ω). (9) If function F (ω) is analytic, the integral in Eq.(6) can be done by closing the contour in the complex plane. F (t) for t will be then given by an exponentially decaying function exp( ω mint), where ω min is the imaginary part of that pole of F (ω) which is closest to the real axis. For example, if F (ω) = (ω 2 + a 2 ), F (t) exp ( at). Thus, the large-t asymptotes of analytic functions decay exponentially in time. On the other hand, if F (ω) is non-analytic, F (t) decays much slower only as a power-law. For example, for F (ω) = exp ( a ω ), we have dω F (t) = 2π e iωt exp ( a ω ) = = 2Re 0 dω 2π e iωt e aω = 2Re 2π 0 dω ( e iωt + e iωt) e aω 2π a + it = π a a 2 + t for t. 2 t2 In addition, if F (ω) has a divergent derivative of order n at finite ω, e.g., for ω = ω 0, that F (t) oscillates in t. This can be seen by doing the partial integration in Eq.(6) n + times ω0 2πF (t) = dωe iωt F (ω) = dωe iωt F (ω) + dωe iωt F (ω) = ω 0 7

8 = iω e iωt F (ω) ω 0 + ( ) iω e iωt F (ω) ω 0 iω dωe iωt d dω F (ω) +..., i.e., until the boundary terms gives the divergent expression dn dω n F (ω 0 ) e iω 0t which oscillates as e iω 0t. 2. End of discourse Coming back to Eq.(9), we can now understand why the induced density around the point charge oscillates as cos 2k F r and falls off only as a power law of the distance ρ ind cos 2k F r r 3. Both of these effects are the consequences of the singularity of ɛ (q, 0) at q = 2k F. E. Friedel oscillations The physics of Friedel oscillations is very simple: they arise due to standing waves formed as a result of interference between incoming and backscattered electron waves. For the sake of simplicity, let s analyze a D case. Suppose that at x = 0, we have an infinitely high barrier (wall). For each k, the wavefunction is a superposition of the incoming plane wave L /2 e ikx and a reflected wave L /2 e ikx : Ψ = L /2 e ikx L /2 e ikx = 2i sin kx L/2 The probability density Ψ 2 = (4/L) sin 2 kx oscillates in space. If the probability that the state with momentum k is occupied is a smooth function of k (as it is the case for the Maxwell-Boltzmann or Bose-Einstein statistics), then summation over k would smear out the oscillations. However, for the Fermi statistics, f k has a sharp (at T = 0) boundary between the occupied and empty states. summation of k. The profile of the density is described by n (x) = 2 k dk 2π f k Ψ 2 = 8 = n 0 sin 2k F x, πx kf 0 As a result, oscillations survive even after the dk 2π sin2 kx = 4 kf 0 dk ( cos 2kx) 2π where n 0 = 2k F /π is the density of the homogeneous electron gas. Away from the barrier, oscillations die off as x. At the barrier, n (0) = 0. 8

9 A 3D case, is different in that the x decay changes to a r 3 one. (In general D- dimensional case, the Friedel oscillations fall off as r D.) Friedel oscillations were observed in STM experiment (see attached figures).. Enhancement of the backscattering probability due to Friedel oscillations As it was discussed in the previous Section, Friedel oscillations arise already in the singleparticle picture. However, they influence scattering of interacting electrons at impurities and other imperfections. Once a Friedel oscillation is formed, the effective potential barrier seen by other electrons is the sum of the bare potential plus the potential produced by the Friedel oscillation. Consider a simple example when D electrons interact via a contact potential U (x) = uδ (x). The potential produced by the Friedel oscillation is V F (x) = dx (n (x ) n 0 ) V (x x ) = u (n (x) n 0 ) = u sin 2k F x. πx Backscattering at an oscillatory potential is enhanced due to resonance. In the Born approximation, the backscattering amplitude for an electron with momentum k is A = dx ( e ikx) VF (x) e ikx = u dx π 0 x sin 2k F xe i2kx = u dx ( e 2i(k+k F )x + e ) 2i(k k F )x. π x 0 The first term gives a convergent integral (we remind that our k > 0), so forget about it. (It s only role is to guarantee the convergence at x 0, but we will take this into account by cutting the integral at x k F..)The second term becomes log-divergent at large distances if k = k F. To estimate the integral, notice that it diverges for k = k F k k F. Thus A = u π k kf k F dx x = u π ln k F k k F. and converges for Precisely at the Fermi surface (k = k F ), the backscattering amplitude (and thus the probability) blows up which means that impurity becomes impenetrable. This effect in D is usually described as arising due to the non-fermi-liquid nature of the ground state?. However, as we just saw that this effect can be simply understood in terms of backscattering from the Friedel oscillations. Higher orders in the e-e interaction can be summed up (see 9

10 Ref. ). It turns out that each next order in u brings in an additional log of k k F. The transmission amplitude of the barrier becomes k F t = t 0 ( g ln k k F + 2 g2 ln 2 k F k k F ) 6 g3 ln 3 k F k k F +... ( ) ( ) g k F k kf = t 0 exp g ln = t 0, k k F k F where g = u/πv F is the dimensionless coupling constant and t 0 is the transmission amplitude in the absence of e e interaction. At k = k F, the transmission amplitude vanishes. Suppose one measures a tunneling conductance of the barrier inserted into a D system. Then typical k k F max {T/v F, ev/v F }, where V is the applied voltage. With the help of the Landauer formula G = (2e 2 /h) t 0 2 we conclude that the tunneling conductance exhibits a power-law scaling in the voltage or temperature G (max {T, ev }) 2g. Such a power-law scaling was indeed observed in carbon nanotubes (which are essentially quantum wires with two channels) F. Hamiltonian of the jellium model The model system we study are electrons in the presence of a positively charged ions. The ionic charge is assumed to be spread uniformly over the system volume ( jellium model ). The net electron charge is equal in magnitude and opposite in sign to the that of ions so that overall the system is electro-neutral. This part of the model is common for all tractable models of e-e interactions in solids, including but not limited to the Hartree-Fock approximation. The classical energy of the system electrons + ions is E int = d 3 r d 3 r 2 n ( r ) V ee ( r r 2 ) n ( r 2 ) + d 3 r d 3 r 2 n ( r ) V ei ( r r 2 ) n i (0) 2 + d 3 r d 3 r 2 n i V ii ( r r 2 ) n i, 2 where n ( r) is the (non-uniform) electron density, n i is the (uniform) density of ions, V ee,ei,ii are the potentials of electron-electron, electron-ion, and ion-ion interactions. In the Hartree- Fock approximation, V ee ( r r 2 ) = e 2 r r 2 = V ei ( r r 2 ) = V ii ( r r 2 ). () 0

11 Using (), Eq.(2) can be re-arranged as E int = d 3 r d 3 r 2 [n ( r ) n i ] V ee ( r r 2 ) [n ( r 2 ) n i ]. 2 In this form, the energy has a very simple meaning. At point r the electron density deviates from the ionic density, so that the net charge density is n ( r ) n i. Similarly, at point r 2 the net charge density is n ( r 2 ) n i. These local fluctuations in density interact via the Coulomb potential. In the absence of any external perturbations, boundaries, and impurities, the average electron density is uniform and equal to the density of ions: n ( r ) = n i. However, the ground state energy involves the product of densities at different points n ( r ) n ( r 2 ) (a correlation function) which is not uniform. In classical systems, local fluctuations in the charge density are due to thermal motion of electrons. In a quantum-mechanical system at T = 0 they are due to the zero-point motion of electrons (in a Fermi system, the kinetic energy is finite at T = 0). To treat the QM system, we need to pass from the classical energy to a Hamiltonian E int Ĥint = d 3 r d 3 r 2 [ˆn ( r ) n i ] V ee ( r r 2 ) [ˆn ( r 2 ) n i ], (2) 2 where now ˆn ( r) is a number-density operator. Densities of ions do not fluctuate so we can leave them as classical variables (c-numbers). Also, the potential is the same as in the classical system. It is convenient to re-write Eq. (2) in a Fourier-transformed form Ĥ int = [ˆn n 2 L 3 i ] q V ee (q) [ˆn n i ] q, q where ˆn ( r) n i = [ˆn ( r) n L 3 i ] q e i q r q V ee ( r) = V L 3 ee (q) e i q r, q with V ee (q) = 4πe 2 /q 2 and where we took into account that V ee (q) depends only on the magnitude of q. The Fourier transform of a uniform density of ions is n i L 3 δ q,0 so Ĥ int = ] [ˆn q n 2 L 3 i L 3 δ q,0 Vee (q) [ˆn ] q n i L 3 δ q,0 q = [ˆn 2 L 3 q Nδ q,0 ] V ee (q) [ˆn q Nδ q,0 ], q

12 where N is the total number of ions (equal to that of electrons). On the other hand, ˆn q=0 = d 3 rˆn ( r) = ˆN, where ˆN is an operator of the total number of particles. Because the total number of particles is fixed this operator is simply equal to its expectation value N. Thus the q = 0 term gives no contribution to the sum. Physically, it means that the fluctuation of infinite size in real space (corresponding to q = 0) do not interact as they are compensated by uniform charge of ions. With that, we re-write Ĥint as Ĥ int = ˆn 2 L 3 q V ee (q) ˆn q. (3) At T = 0, the ground-state energy is simply an expectation value of Ĥint : E int = 0 Ĥint q 0 0 = V 2 L 3 ee (q) 0 ˆn qˆn q 0. As we simply do the perturbation theory with respect to V ee, the expectation value is calculated using the wave-functions of the free system (vacuum average). Now we need the second quantized from of ˆn q. To this end, we recall that in real space q 0 ˆn ( r) = α Ψ α ( r) Ψ α ( r). Operator Ψ α ( r) (Ψ α ( r)) creates (annihilates) a particle with spin projection α at point r. Using the plane waves as our basis set, Ψ α ( r) = c L 3/2 pα e ip r Ψ α ( r) = c L pαe ip r. 3/2 ˆn q = d 3 re i q rˆn ( r) = d 3 re i q r Ψ α ( r) Ψ α ( r) α = d 3 re i q r c eip r L 3 p α c pα e ip r α p,p = c p q,α c pα. α p 2 p p

13 Substituting this last result into Eq.(3), we obtain Ĥ int = V 2 L 3 ee (q) c p q,α c pαc k+ q,β c kβ. q 0 p,k α,β It is convenient to re-write the second-quantized Hamiltonians in the normal-ordered form, when all creation operators are positioned to the right of the annihilation ones. Interchanging the positions of two fermionic operators twice (no sign change!) and discarding the terms which result from the δ-function part of the anti-commutation relations (those will give only a shift of the chemical potential), we arrive at Thus, E int = 0 Ĥ int = V 2 L 3 ee (q) c p q,α c k+ q,β c kβc pα. Ĥint q 0 p,k α,β 0 = V 2 L 3 ee (q) 0 c p q,α c k+ q,β c kβc pα 0. q 0 p,k α,β The expectation value of the product of c operators can be calculated directly. However, it is much more convenient to use the result known as the Wick s theorem 2, 3. The Wick theorem states that an expectation value of a product of any number of c operators splits into products of expectation values of pairwise averages (called contractions): 0 Π M i= c p i,α i Π M 2 j= c p j,α j 0 = P Π M i=p i,α i Π M 2=M j= F ij 0 c pi,α i c pj,α j 0 for M = M 2 and is equal to zero otherwise. The sum goes over permutations. The factor F ij is equal to one if it takes an even number of permutations to bring c p i,α i c pj,α j together (no permutation is an even permutation of order zero) and it is equal to minus one if takes an odd number of permutations. The The pairwise averages are nothing more then occupation numbers where we suppress the spin index of f pi 0 c pi,α i c pj,α j 0 = δαi α j δ pi,p j f pi. For our case, we can pair operators in two ways assuming that the ground state is paramagnetic. 0 c p q,α c k+ q,β c kβc pα 0 = 0 c p q,α c pα 0 0 c k+ q,β c kβ0 0 c p q,α c kβ 0 0 c k+ q,β c pα 0 = δ q,0 f p f k δ αβ δ q,p k f p f k. 3

14 The first term drops out from the sum, whereas the second one gives E int = V 2 L 3 ee ( p k ) f p f k. p k α The effective single-particle energy can be found as a variational derivative of the ground state energy ε p = δ δf p (E 0 + E int ) = p2 p2 2m + δε p. 2m L 3 Calculate the integral over k at T = 0 where y p/k F. k V ( k k )f k = p2 2m δε p (p) = kf π e2 dkk 2 d cos θ p 2 + k 2 2pk cos θ = π e 2 p = e2 k F π 0 kf dkk ln p + k 0 p k ( + ) y2 ln + y 2y y, Near the bottom of the band when p k F, i.e., y, we have Σ (p) = e2 k F (2 23 ) π y2 = 2e2 k F + 2e2 p 2. π 3πk F Adding this up with the free spectrum, we obtain ε p = 2e2 k F π + p2 2m + 2e2 p 2 3πk F = 2e2 k F π + p2 2m. d 3 k (2π) 3 V ( k k )f k A constant term means that the chemical potential is shifted by the interaction. The effect we are after is the change of the coefficient in front of p 2 : m = ( ) + 4e2 m m 3πk F or m = m + e2 m 3πk F = m r s. Near the bottom of the band, the effective mass of electrons is smaller than the band mass. However, this is not the part of the spectrum we are interested in thermodynamic and transport phenomena. In this phenomena, the vicinity of k F (4) (5) plays the major role. Away from the bottom of the band the self-energy is not a quadratic function of p, so the resulting spectrum is not parabolic (see Fig.??). We need to understand what is the meaning of the effective mass in this situation. 4

15 G. Effective mass near the Fermi level In a Fermi gas, the Fermi momentum is obtained by requiring that the number of states within the Fermi sphere is equal to the number of electrons: πk3 F (2π) 3 = n. This condition does not change in the presence of the interaction. Therefore, k F of the interacting system is the same as in a free one. (This statement can be proven rigorously and is known as the Luttinger theorem 3.) The energy of a topmost state is the chemical potential. In a Fermi gas, µ = ε pf = p2 F 2m. When the spectrum is renormalized by the interaction, µ = ε p F. The energy of a quasi-particle is defined as ɛ p = ε p µ. From the definition of the self-energy, ɛ p = ε p µ = ε p + Σ (p) µ. At the Fermi surface, this equality reduces to 0 = p2 F 2m + Σ (p F ) µ µ = p2 F 2m + Σ (p F ). The chemical potential is changed due to the interaction. Near the Fermi surface, i.e., for p p F p F, ɛ p = p2 2m + Σ (p p F + p F ) p2 F 2m Σ (p F ). Near the Fermi surface, the spectrum can always be linearized p 2 2m p2 F 2m = (p p F ) (p + p F ) 2m 5 v F (p p F ).

16 Expand the self-energy in Taylor series Σ (p p F + p F ) Σ (p F ) Σ (p p F ), where Σ Σ/ p p=pf. The renormalized spectrum can be also linearized near the Fermi surface ɛ p v F (p p F ), where v F is the renormalized Fermi velocity. Now we have v F (p p F ) = v F (p p F ) + Σ (p p F ) v F = v F + Σ The effective mass near the Fermi surface is defined as so that or m v F p F m = m + Σ p F m = m. (6) + Σ /v F m is determined by the derivative of the self-energy at the Fermi surface.. Effective mass in the Hartree-Fock approximation Calculating the derivative of the self-energy in the Hartree-Fock approximation [Eq.(5)], we arrive at an unpleasant surprise: Σ =. According to Eq.(6), this means that m = 0. This unphysical result is the penalty we pay for using an oversimplified model in which electrons interact via the unscreened Coulom potential. 2. Beyond the Hartree-Fock approximation The deficiency of the Hartree-Fock approximation is cured by using the screened Coulomb potential. Because electrons exchange energies, the polarization clouds of induced charges around them are dynamic, i.e., the two-body potential depends not only on the distance but 6

17 also on time. In the Fourier space, it means the interaction is a function not only of q but also of frequency ω V ee (q, ω) = 4πe2 q 2 ɛ (q, ω), where ɛ (q, ω) is the full Lindhard function. A calculation of the self-energy in this case is a rather arduous task, so I will give here only the result for the effective mass near the Fermi level 3 m = e2 ln 2p F πv F eκ, where e = where κ is the screening wavevector In terms of r s. m κ 2 = 4πe 2 ν F. m m = ( ) ( /3 4 r s ln 2π 9π π (2.78) 2 ( ) ) /3 4 9π r s =.0 8r s ln 0.22 r s. This expression is valid for a weak interaction, i.e., r s. For very small r s, m decreases with r s. At r min s 0.08, m has a minimum and it becomes equal to m at r s III. STONER MODEL OF FERROMAGNETISM IN ITINERANT SYSTEMS Consider a model of fermions interacting via a delta-function potential V ( r r 2 ) = gδ ( r r 2 ) (a good model for He 3 atoms). The interaction part of Hamiltonian reads (in real space) H int = d 3 r d 3 r 2 V ( r r 2 ) Ψ α ( r ) Ψ β 2 ( r 2) Ψ β ( r 2 ) Ψ α ( r ). For a delta-function interaction, we need to take into account that Ψ α ( r) Ψ β ( r) 0 only if α β. H int then reduces to H int = g d 3 rψ ( r) Ψ ( r) Ψ ( r) Ψ ( r) An expectation value of the interaction energy per unit volume is given by E int = L 3 0 H int 0 = L 3 g d 3 r 0 Ψ ( r) Ψ ( r) Ψ ( r) Ψ ( r) 0 (7) (using Wick s theorem) = L 3 g d 3 r 0 Ψ ( r) Ψ ( r) 0 0 Ψ ( r) Ψ ( r) 0 (8) = L 3 g d 3 rn n = gn n (9) 7

18 where n, is the (expectation value of) number density of spin-up (down) fermions, which is position independent. In a paramagnetic state, n = n = n/2, where n is the total number density. Let s analyze the possibility of a transition into a state with a finite spin polarization (ferromagnetic state), in which n n. The kinetic energy of a partially spin-polarized system is: E 0 = 0<k<k F d 3 k (2π) 3 ε k + 0<k<k F d 3 k (2π) 3 ε k, where k, F are the Fermi momenta of the spin-up (down) fermions related to n, via 0<k<k α F 4 3 π (k, F )3 (2π) 3 d 3 k (2π) 3 ε k = = n, E 0 = (6π2 ) 5/3 20π 2 m k, F = ( 6π 2) /3 n /3, 4π k, F (2π) 3 dkk 2 k2 0 [ ] n 5/3 + n 5/3 F )5 2m = (k, 20π 2 m Introducing the full density n = n + n and the difference in densities δn = n n, so that n, = (/2) (n ± δn), the equation for E 0 takes the form E 0 = (6π2 ) 5/3 [ (n + δn) 5/3 + (n δn) 5/3]. 20π 2 2 5/3 m Obviously, E 0 has a minimum at δn = 0, i.e., in a paramagnetic state. The interaction term changes the balance: E int = gn n = (g/4) ( n 2 (δn) 2). For a repulsive interaction (g > 0), the interaction part of the energy is lowered by spin polarization. The total energy is the sum of two contributions E = E 0 + E int = (6π2 ) 5/3 20π 2 2 5/3 m Introducing dimensionless quantity [ (n + δn) 5/3 + (n δn) 5/3] + (g/4) ( n 2 (δn) 2). ζ = δn/n 8

19 and using the relation between the Fermi energy and the density, the expression for energy is simplified to E = 3 ] 0 ne F [( + ζ) 5/3 + ( ζ) 5/3 + ( n 2 g/4 ) ( ζ 2). Suppose that ζ is small, then the first term can be expanded to the second order (the first order contribution vanishes), using ( + ζ) 5/3 + ( ζ) 5/3 = ζ ζ4 + O (x 5 ) : E (ζ) = 3 5 ne F + 3 ne F ζ 2 + ( n 2 g/4 ) ( ζ 2) + 8 gne F ζ where E (0) = 3 5 ne F + n 2 g/4 and = E (0) + aζ gne F ζ , a = 3 ne F ( 3g ) n = ( 4E F 3 ne F gν () F with ν () F = 3g/4E F being the density of states per one spin orientation. If a > 0, the ferromagnetic state is energetically unfavorable. When it is negative, the the ferromagnetic state is energetically unfavorable. The transition occurs when the coefficient vanishes. Thus the critical value of the coupling constant is g c = /ν () F. For g > g c, polarization is stabilized by the quartic term. An equilibrium value of polarization is determined from the condition E ζ = 0 ) 2 3 ne F ( ggc ζ g cne F ζ 3 = 0 ζ = ( ) /2 g gc 27/2. Notice that we replace g by g c in the quartic term, which is justified near the critical point. Unfortunately, the critical value for g is outside the weak coupling regime and, therefore, the Stoner model cannot be considered as a quantitative theory of ferromagnetism. However, it does provide with a hint as to how ferromagnetism occurs in real system. The Stoner model is also probably the first example of zero-temperature (or quantum) phase transitions, i.e., phase changes in the ground state of the system driven by varying some control parameter 9 g c ),

20 (in this case, the effective interaction strength). Incidentally, it is also a second-order or continuous phase transition: the energy of the system is continuous through the critical point. The equilibrium value of the order parameter ( ζ) is zero for g < g c and becomes finite but still small for g > g c. A square-root dependence of ζ on the devitation of the control parameter from its critical value is a characteristic feature of the mean-field theories. The Stoner model also predicts that the spin susceptibility of the paramagnetic state is enhanced by the repulsive interaction. To this end we need to introduce magnetization (the magnetic moment per unit volume) M = g L µ B δn = g L µ B nζ, where g L is the Lande g factor (not to be confused with the coupling constant!) and µ B is the Bohr magneton, and switch from the ground state energy to the free energy (at T = 0) F = E MH. Retaining only quadratic in M terms, we have F = E (0) + a (gµ B n) 2 M 2 MH. An equlibrium magnetization is obtained from the condition F M = 0 M = (gµ Bn) 2 H. 2a Recalling that the spin susceptibility is defined by the following relation, we read off the susceptibility as χ = (g Lµ B n) 2 2a (g L µ B n) 2 = ( ) (2/3)nE F gν () F = (g Lµ B ) 2 ν F gν () F χ 0 gν () F = Sχ 0. where ν F is the density of states per two spin orientations, χ 0 is the spin susceptibility of a free electron gas at T = 0 (Pauli susceptibility), and S gν () F 20

21 is called the Stoner enhancement factor. For g > 0, χ > χ 0. Also, χ diverges at the critical point. A divergent susceptibility is another characteristic feature of both finite- and zerotemperature phase transitions. What is the magnetic response of the system at the critical point, where the (linear) susceptibility χ =? To answer this, we need to restore the quartic term in the free energy a F = E (0) + (g L µ B n) 2 M 2 MH + ge F 8 (g L µ B ) 2 n M 4. At the critical point, g = g c and a = 0. The equation of state then reads F M = H + 4 ge F 8 (g L µ B ) 2 n M 3 = 0. Therefore, at the critical point M scales as H /3. Note that a magnetic field smears the phase transition because now there is a finite magnetization even above the critical point. The order parameter now changes continuously through the transition which is defined as a critical value of g, where the linear susceptibility diverges. IV. WIGNER CRYSTAL As we now understand the properties of a weakly interacting electron gas, let s turn to the opposite limit, when the interaction energy is much larger than the kinetic one. T = 0, the kinetic energy is the Fermi energy so the condition for the strong interaction is r s. What happens in this limit? The answer was given by Wigner back in 934. Because the Coulomb energy is very high, electrons would try to be as further away from each other as possible. Ideally, they would all move to the sample boundaries. However, this would create an enormous uncompensated positive charge of ions. The next best thing is to arrange into an electron lattice (Wigner crystal) of spacing comparable to the average inter-electron distance in the liquid phase. Let s estimate when the formation of the Wigner crystal is possible. In a liquid, come to each other at arbitrarily small distances, where the Coulomb energy is large. In a crystal, electrons are separated by the lattice spacing, a. The gain in the potential energy P L P C e2 a (for the sake of simplicity I assume that the dielectric constants of ions is unity). On the other hand, kinetic energy in a liquid is K L /ma 2 whereas that in a crystal K C /mr 2 0, 2 At

22 where r 0 a is the rms displacement of an electron about its equilibrium position due to the zero-point motion. For the crystal to be stable, one must require that r 0 a. Thus K C K L and K C K L /mr 2 0. Crystallization is energetically favorable when the gain in the potential energy exceeds the loss in the kinetic one, i.e., or P L P C K C K L e 2 a /mr2 0. (20) Now I want to show that the condition above is nothing is equivalent to r s. To estimate r 0, consider an oscillatory motion of an electron in a D lattice interacting with its nearest neighbors via Coulomb forces. The potential energy of the central electron when it is moved by distance x from its equilibrium position (x = 0) is U (x) = e2 a x + e2 a + x. Expanding the expression above for x a, we obtain e 2 U (x) = 2e2 a + 2 a 3 x2. The harmonic part of the potential is reduced to the canonical form by equating e 2 2 a 3 x2 2 mω2 0x 2 ω 2 0 = e2 ma 3. Notice that because a 3 n, ω 0 is of order of the plasma frequency 4πne 2 /m which is quite a natural result. Thus the Debye frequency of a Wigner crystal is the plasma frequency. The quantum amplitude is related to the frequency via mr 2 0 ω 0 or ( mr 2 0 ) 2 ω 2 0 e2 ma 3. (2) 22

23 Squaring Eq.(40) and using Eq.(2), we find ( e 2 a ) 2 ( ) /mr0 2 2 e 2 = ma 3 e 2 ma 2 or r s. Thus crystallization is energetically favorable for r s. Quantum Monte Carlo simulations show that the critical value of r s, at which the liquid crystallizes, is 50 in 3D and 37 in 2D. Why so huge numbers? This one can understand by recalling that crystals melt when the amplitude of the oscillations is still smaller than the lattice spacing. The critical ratio of the amplitude to spacing is called the Lindemann parameter Λ. For all lattices, Λ is appreciably smaller than unity (this helps to understand why the melting temperatures are significantly smaller than cohesive energies). Typically, Λ (Strictly speaking, one has to distinguish between classical and quantum Lindemann parameters since entropy plays role for the former but not for the latter but we will ignore this subtlety). With the help of Eq.(2), we find that Wigner crystal melts when r 0 /a = r /4 s. r 0 /a = Λ r c s = Λ 4. Wigner crystals were observed in layers of electrons adsorbed on a surface of liquid helium (that makes the smoothest substrate one can think of). The search for Wigner crystallization in semiconductor heterostructures is a very active field which so far has not provided a direct evidence for this effect although a circumstantial evidence does exist. The main problem here is that effects of disorder in solid-state structures becomes very pronounced at lower densities so there is no hope to observe Wigner crystallization in its pure form. What one can hope for is to get a distorted crystal (Coulomb glass). 23

24 V. FERMI-LIQUID THEORY A. General concepts A detailed theory of the Fermi liquid (FL), and its microscopic justification in particular, goes far beyond the scope of this course. Standard references 3, 4, 5 provide an exhaustive if not elementary treatment of the FL theory. In what follows, I will do a simplified version of the theory ( FL-lite ) and illustrate main concepts on various examples.. Motivation All Fermi systems (metals, degenerate semiconductors, normal He 3, neutron stars, etc.) belong to the categories of either moderately or strongly interacting systems. For example, in metals r s in the range from 2 to 5. (There are only few exceptions of this rule; for example, bismuth, in which the large value of the background dielectric constant brings the value of r s to 0.3 and GaAs heterostructures in which the small value of the effective mass 0.07 of the bare mass leads to the higher value of the Fermi energy and thus to r s < in a certain density range). On the other hand, as we learned from the section on Wigner crystallization, the critical r s for Wigner crystallization is very high 50 in 3D and 37 in 2D. Thus almost all Fermi systems occuring in Nature are too strongly interacting to be described by the weak-coupling theory (Hartree-Fock and its improved versions) but too weakly interacting to solidify. In short, since they are neither gases nor solids the only choice left is that they are liquids. A liquid is a system of interacting particles which preserves all symmetries of the gas. Following this analogy, Landau put forward a hypothesis that an interacting Fermi system is qualitatively similar to the Fermi gas 6. Although original Landau s formulation refers to a translationally invariant system of particles interacting via short-range forces, e.g., normal He 3, later on his arguments were extended to metals (which have only discrete symmetries) and to charged particles. Experiment gives a strong justification to this hypothesis. The specific heat of almost all fermionic systems (in solids, one need to subtract off the lattice contribution to get the one from electrons) scales linearly with temperature: C (T ) = γ T. (Some systems demonstrate the deviation from this law and these systems are subject of an active studies for the last 0 years; see more in Non-fermi-liquid behavior). In a free Fermi gas, γ = γ = (π 2 /3) ν F = 24

25 (/3) mk F. In a band model, when non-interacting electrons move in the presence of a periodic potential), one should use the appropriate value of the density of states at the Fermi level for a given lattice structure. In reality, the coefficient γ can differ significantly from the band value but the linearity of C(T ) in T is well-preserved. In those cases, when one can change continously the interaction (for example, by applying pressure to normal He 3 ), γ is found to vary. One is then tempted to assume that the interacting Fermi liquid is composed of some effective particles (quasi-particles) that behave as free fermions albeit their masses are different from the non-interacting values. B. Scattering rate in an interacting Fermi system The Pauli principle leads to a slow-down of mutual scattering of fermions in a degenerate Fermi system. A qualitative argument is that two fermions can interact effectively if their energies happen to be within the T intervals around the Fermi energy. that one the energies is within this interval is of order T/E F The probability and, since the particles are independent, the scattering probability is proportional to (T/E F ) 2 which is much smaller than unity for T E F. A precise definition of the scattering rate depends on the quantity measured. In general, different quantities, such as charge and thermal conductivities, contain scattering rates that differ at least by numerical factors. To avoid this complication, we adopt one of the possible definitions of the scattering rate. Consider a scattering process in which an electron with momentum k collides with an electron with momentum k such that the momenta of the final states are k q and k + q, correspondingly. The interaction is assumed to be a screened Coulomb potential U(q, ω) = 4πe2 q 2 ɛ(q, ω), (22) where ɛ(q, ω) is the dielectric function of the electron gas. The Fermi golden rule for the number of transitions from a given state k is τ = 2π d 3 p (2π) 3 d 3 q (2π) 3 U(q, ε k ε k q ) 2 δ(ε k +ε k ε k q ε k+q )f 0k ( f 0k q )( f 0k+q ), where f 0k f 0 (ε k ) is the Fermi function. The combination of the Fermi functions in the equation above ensures that the initial state k of the scattering process is occupied while 25 (23)

26 the final ones are empty. It is convenient to introduce the energy transfer ω = ε k ε k q = ε k+q ε k and re-write the δ-function as an integral over ω: δ(ε k + ε k ε k q ε k+q ) = dωδ(ε k ε k q ω)δ(ε k+q ε k ω). Then τ = 2π d 3 p d 3 q dω U(q, ω) 2 δ(ε (2π) 3 (2π) 3 k ε k q ω)δ(ε k+q ε k ω) f 0 (ε k )( f 0 (ε k ω))( f 0 (ε k + ω)), (24) We assume now (and justify later) that ɛ(q, ω) can be taken in the static limit (ω = 0) and that typical q k F, such that we can use the Thomas-Fermi result for ɛ(q k F, 0) = + κ 2 /q 2 with κ 2 = 4πe 2 ν(e F ): U(q, ω) U(q, 0) = 4πe2 q 2 + κ 2. (25) We will see that once the interaction potential is assumed to be static, the T 2 form of /τ is obtained under very broad assumptions on the potential. We choose k to be close to k F and assume that p is near k F as well. The arguments of the δ functions in q can be rewritten as ( k 2 (k q)2 δ(ε k ε k q ω) = δ 2m 2m ( (k + q) 2 δ(ε k+q ε k ω) = δ 2m ) ω ) k2 2m ω ( kq = δ = δ ) 2m ω ) m cos θ kq q2 ( pq m cos θ pq + q2 2m ω, (26) where θ kq = {k, q} and θ kq = {k, q}. The energy transfers will be shown to be small: of order T. Therefore, we can neglect ω in the arguments of the δ functions (but keep it in the arguments of the Fermi functions because there ω is divided by T ). The angular integrations can now be readily performed. Write the angular parts of the integral as d(cos θkq ) d(cos θ pq ) and integrate in this particular order (that is, first integrate over the direction of k at fixed q and then integrate over the direction of q at fixed k). With ω = 0 in the δ-functions, the values of the cosines are cos θ kq = q/2k q/2k F (since by assumption q k F and cos θ pq = q/2p q/2k F, which is well within the integration interval < cos θ kq, cos θ pq <. Performing angular integration, we obtain τ = dpp 2 dq dω U(q, 0) 2 32π 3 vf 2 0 f 0 (ε k )( f 0 (ε k ω))( f 0 (ε k + ω)), (27) Now, taking into account that p is near k F, we approximate 0 dpp 2 as d(p k F )k 2 F = k 2 F vf dξ p where ξ p = v F (p k F ) is the energy of excitation measured from the Fermi 26

27 energy. Since the Fermi functions contain only the differences ε k E F ξ k, etc., we can replace ε k by ξ k and ε k by ξ p in there: τ = k2 F 32π 3 v 3 F 0 dq U(q, 0) 2 dξ p dωf 0 (ξ p )( ( f 0 (ξ p + ω)). (28) We now see that the energy and momentum integrals are completely decoupled. First we integrate over ξ p using dξ p f 0 (ξ p )( f 0 (ξ p + ω)) = Rescaling ω = yt, the remaining integral is reduced to where F (x) = τ = T 2 m 2 32π 3 v F For a particle right on the Fermi surface (ξ k = 0), we obtain 0 ω exp( ω/t ) (29) dq U(q, 0) 2 F (ξ k /T ), (30) y dy (exp(y x) + )( exp( y)). (3) F (0) = 0 y dy sinh y = π2 4 (32) and τ = T 2 m 2 28πv F dq U(q, 0) 2. (33) As we had announced, the T 2 law is independent of a particular form of U(q, 0). For a screened Coulomb potential, we obtain 0 dq U(q, 0) 2 = (4πe 2 ) 2 dq (q 2 + κ 2 ) = (4πe2 ) 2 π 2 κ 3 4 = πκ 4ν 2 (E F ). (34) Recalling that ν(e F ) = k 2 F /π2 v F, the final result can be written as or, on restoring k B and, 0 τ = τ = π4 κ T 2 (35) 024 k F E F π4 κ (k B T ) 2. (36) 024 k F E F Now, we need to verify assumptions made en route to the final result. We assumed that ω can be set to zero in the dielectric function and that typical q k F. From the last integral over q, we see that typical q κ k F by assumption of weak e-e interaction, hence the second assumption is satisfied. On the other hand, typical values of y in (32) are, hence 27

28 typical ω T. The dielectric function depends on the combination ω/v F q. hence the static limit is justified if ω v F q v F κ. Notice that vf 2 κ2 vf 2 e2 kf 2 /v F e 2 n/m ωp, 2 where ω p E F is the plasma frequency. Therefore, the T 2 law in an electron system is valid for T ω p. The integral in Eq. (3) can be calculated exactly with the result F (x) = 2 (x2 + π 2 ). (37) + e x Then we get instead of Eq. (35) at finite ξ k τ = π2 52 κ ξk 2 + π2 T 2 k F E F + exp( ξ k /T ) (38) C. Quasi-particles The concept of quasi-particles central to the Landau s theory of Fermi liquids. The ground state of a Fermi gas is a completely filled Fermi sphere. The spectrum of excited states can be classified in terms of how many fermions were promoted from states below the Fermi surface to the ones above. For example, the first excited state is the one with an electron above the Fermi sphere and the hole below. The energy of this state, measured from the ground state, is ɛ = p 2 /2m E F. The net momentum of the system is p. The next state correspond to two fermions above the Fermi sphere, etc. If the net momentum of the system is p, then p + p 2, where p and p 2 are the momenta of individual electrons. We see that in a free system, any excited macrosopic state is a superposition of single-particle states. This is not so in an interacting system. Even if we promote only one particle to a state above the Fermi surface, the energy of this state would not be equal to p 2 /2m E F because the interaction will change the energy of all other fermions. However, Landau assumed that excited states with energies near the Fermi level, that is, weakly excited states, can be described as a superposition of elementary excitations which behave as free particles, although the original system may as well be a strongly interacting one. An example of such a behavior are familiar phonons in a solid. Suppose that we have a gas of sodium atoms (which are fermions) which essentially don t interact because of low density. The elementary excitations in an ideal gas simply coincide with real atoms. Now we condense gas into metals. Individual atoms are not free to move on their own. Instead, they can only 28

29 participate in a collective oscillatory motion which is a sound wave. For small frequencies, the sound wave can be thought of consisting of elementary quanta of free particles phonons. The spectrum of each phonon branch is ω i (q) = s i q, where s i is the speed of sound and the oscillatory energy is E = i d 3 q (2π) 3 ω in i, where n i is the number of excited phonons at given temperature. If the number of phonons is varied, so is the total energy δe = i d 3 q (2π) 3 ω iδn i. Quite similarly, Landau assumed that the variation of the total energy of a single-component Fermi liquid (or single band metal) can be written as d 3 p δe = 3 ε (p) δn (p). (39) (2π) (For the sake of simplicity, I also assume that the system is isotropic, i.e., the energy and n depend only on the magnitude but not the direction of the momentum but the argument extends easily to anisotropic systems as well). In this formula, n (p) is the distribution function of quasi-particles which are elementary excitations of the interacting system. In analogy with bosons, these quasi-particles are free. One more and crucial assumption is that the quasi-particles of an interacting Fermi system are fermions which is not at all obvious. For example, regardless of the statistics of individual atoms which can be either fermions or bosons, phonons are always bosons. Landau s argument was that if quasiparticles were bosons they could accumulate without a restriction in every quantum state. That means that an excited state of a quantum system has a classical analog. Indeed, an excited state of many coupled oscillators is a classical sound wave. Fermions don t have macrosopic states so quasi-particles of a Fermi systems must be fermions (to be precize they should not be bosons; proposal for particles of a statistics intermediate between bosons and fermions anyons have been made recently). Thus, n = exp ( ) ε µ, T + On quite general grounds, one can show that quasi-particles must have spin /2 regardless of (half-integer) spin of original particles 3, 4. (Thus, quasi-particles of a system composed 29

30 of fermions with spin S = 3/2 still have spin /2.) Generally speaking, the quasiparticle energy is an operator in the spin space: ε ˆε and, correspondingly, n ˆn = exp ( ), (40) ˆε µ T + where ˆε and ˆn are 2 2 matrices. If the system is not in the presence of the magnetic field and not ferromagnetic, ˆε αβ = εδ αβ, ˆn αβ = nδ αβ. In a general case, instead of (39) we have d 3 p δe = 3 Trˆε (p) δˆn (p), (2π) which, for a spin-isotropic liquid, reduces to d 3 p δe = 2 3 Trε (p) δn (p). (2π) The occupation number is normalized by the condition d 3 p d 3 p δn = 3 Trδn (p) = 2 3 δn (p) = 0, (2π) (2π) where N is the total number of real particles. For T=0, the chemical potential coincides with energy of the topmost state µ = ε (p F ) E F. Another important property (known as Luttinger theorem) is that the volume of the Fermi surface is not affected by the interaction. For an isotropic system, this means the Fermi momenta of free and interacting systems are the same. counting of states is not affected by the interaction, i.e., the relation N = 2 4πp3 F /3 (2π) 3 holds in both cases. A general proof of this statement is given in Ref. [ 3 ]. A simple argument is that the. Interaction of quasi-particles Phonons in a solid do not interact only in the first (harmonic) approximation. Anharmonism results in the phonon-phonon interaction. However, the interaction is weak at low 30

31 energies not really because the coupling constant is weak but rather because the scattering rate of phonons on each other is proportional to a high power of their frequency: τ ph-ph ω5. As a result, at small ω phonons are almost free quasi-particles. Something similar happens with the fermions. The nominal interaction may as well be strong. However, because of the Pauli principle, the scattering rate is proportional to (ε E F ) 2 and weakly excited states interact inly weakly. In the Landau theory, the interaction between quasi-particles is introduced via a phenomenological interaction function defined by the proportionality coefficient (more precisely, a kernel) between the variation of the occupation number and the corresponding variation in the quasi-particle spectrum d 3 p δε αβ = (2π) 3 f αγ,βδ (p, p ) δn γδ (p ). (summation over the repeated indices is implied). Function f αγ,βδ (p, p ) describes the interaction between the quasi-particles of momenta p and p (notice that these are both initial states of the of the scattering process). Spin indices α and β correspond to the state of momentum p whereas indices γ and δ correspond to momentum p. In a matrix form, δˆε = Tr d 3 p (2π) 3 ˆf (p, p ) δˆn (p ), (4) where Tr denotes trace over spin indices γ and δ. For a spin-isotropic FL, when δˆε = δ αβ δε and δˆn = δ αβ δn external spin indices (α and β) can also be traced out and Eq.(4) reduces to where d 3 p δε = (2π) 3 f (p, p ) δn (p ), f (p, p ) 2 TrTr ˆf (p, p ). For small deviations from the equilibrium, δn (p ) is peaked near the Fermi surface. Function ˆf (p, p ) can be then estimated directly on the Fermi surface, i.e., for p = p = p F. Then ˆf depends only on the angle between p and p. The spin dependence of ˆf can be established on quite general grounds. In a spin-isotropic FL, ˆf can depend only on the scalar product of spin operators but on the products of the individual spin operators with some other vectors. Thus the most general form of ˆf for a spin-isotropic system is ν ˆf (p, p ) = F s (θ) ÎÎ + F a (θ)ˆσ ˆσ, 3

32 where Î is the unity matrix, ˆσ is the vector of three Pauli matrices, θ is the angle between p and p, and the density of states was introduced just to make functions F s (θ) and F a (θ) dimensionless. (Star in ν means that we have used a renormalized value of the effective mass so that ν = m k F /π 2, but this is again just a matter of convenience.) Explicitly, ν f αγ,βδ (p, p ) = F s (θ)δ αβ δ γδ + F a (θ)ˆσ αβ ˆσ γδ. (42) In general, the interaction function is not known. However, if the interaction is weak, one can relate ˆf to the pair interaction potential. In a microscopic theory, it can be shown (see Appendix A) that if particles interact via a weak pair-wise potential U (q) such that U (0) is finite (which excludes, e.g., a bare Coulomb potential), then to the first order in this potential f αγ,βδ (p, p ) = δ αβ δ γδ [U (0) ] 2 U ( p p ) 2ˆσ αβ ˆσ γδ U ( p p ). On the Fermi surface, p p = 2p F sin θ/2. Comparing this expression with Eq.(42), we find that F s (θ) = ν [ U (0) ] 2 U (2p F sin(θ/2)) F a (θ) = ν 2 U (2p F sin(θ/2)). (43) Notice that repulsive interaction (U > 0) corresponds to the attraction in the spin-exchange channel (F a < 0). D. General strategy of the Fermi-liquid theory One may wonder what one can achieve introducing unknown phenomenological quantities, such as the interaction function or its charge and spin components. FL theory allows one to express general thermodynamic characteristic of a liquid (effective mass, compressibility, spin susceptibility, etc.) via the angular averages of Landau functions F s (θ) and F a (θ). Some quantities depend on the same averages and thus one express such quantities via each other. The relationship between such quantities can be checked by comparison with the experiment. 32